The GeForce RTX 30 sequence is already tremendous quick. The higher-end playing cards in that household chew up and spit out video games at excessive resolutions the best way playing cards from 5 years in the past have been simply beginning to grasp 1080p. It will get slightly bit foolish at occasions after we’re benchmarking these graphics playing cards as a result of it’s generally troublesome to seek out benchmarks that may actually put the harm on one thing like a GeForce RTX 3080 Ti—to say nothing of the 3090 or 3090 Ti.
That’s why speaking concerning the GeForce RTX 4000 sequence can get our heads spinning, as a result of these playing cards are sounding like a monumental leapp forward of the Ampere technology. Most of what we predict we all know concerning the upcoming GeForce technology comes from leakers. We’ve heard tales of large GPUs pulling unimaginable energy numbers, and tall tales of as much as 48GB of reminiscence, however the actuality might be slightly extra modest.
Starting from the highest, the leaker says that the GeForce RTX 4090 will use the AD102-300 GPU, and have absolutely 16,384 FP32 ALUs, or “CUDA cores” in NVIDIA nomenclature. That can be 128 Ada Lovelace Streaming Multiprocessors (SMs), which is an improve from kopite7kimi’s final leak. Previous leaks have indicated that the AD102 chip has 144 SMs in its full-fat configuration, so this leaves some room for a RTX 4090 Ti or “Titan” SKU.
The reminiscence configuration on the RTX 4090 will supposedly be some 24GB of GDDR6X reminiscence operating at a blistering 21Gbps on a 384-bit reminiscence bus. This is, notably, the very same configuration because the GeForce RTX 3090 Ti, which lends some credence to the concept the 2 playing cards (4090 and 3090 Ti) use closely-related PCBs, as has been rumored previously.
Stepping all the way down to the GeForce RTX 4080’s rumored specs, we see a a lot bigger leap than within the 3000 sequence between x80 and x90 playing cards. The GeForce RTX 4080 is supposedly going to ship with a distinct processor altogether in comparison with the top-end card, often called AD103-300. We have seen a “103” GPU earlier than; there is a GA103 that’s used within the top-end Ampere cell GPUs. The Ada Lovelace model will allegedly have 10240 CUDA cores, giving it 80 SMs of a attainable 84.
This newest leak pegs the GeForce RTX 4080 with a narrower 256-bit reminiscence interface utilizing 16GB of GDDR6 reminiscence, probably at 18Gbps. This appears conservative to us; we might anticipate the x80 card to ship with GDDR6X, at the least, and certain on the increased 21Gbps clock fee. The rumored configuration would find yourself with much less reminiscence bandwidth than even the RTX 3070 Ti, though even 256-bit GDDR6X at 21 GT/s can be behind the RTX 3080 and its a lot wider bus. It’s an attention-grabbing rumor for certain; maybe the a lot bigger L2 cache on Ada Lovelace could make up the distinction, or NVIDIA has devised a strategy to successfully enhance reminiscence bandwidth by a brand new compression expertise or Infinity Cache-like characteristic.
Finally, for the RTX 4070, the leak predicts that it’s going to use AD104-275 with 7168 CUDA cores in 56 SMs. Based on the leaked specs for the Ada Lovelace GPUs, that might be the identical 4-SM reduce from the full-fat AD104 GPU as we noticed in AD103 for the RTX 4080. He predicts 10GB of 18Gbps GDDR6 reminiscence on a 160-bit bus for the RTX 4070, one other brief reminiscence bandwidth allocation placing it on par with the RTX 3060. Seems doubtful to us, however once more, Ada Lovelace is drastically growing the on-die GPU cache in comparison with Ampere. Maybe that makes up the distinction.
One sore level of the GeForce RTX household for a lot of PC fanatics has been the sharp rise in worth at each tier. The GeForce GTX 1080 shipped with a nominal MSRP of $599, which dropped to $499 after 9 months. The GeForce RTX 2080 debuted at $699, and the RTX 3080 piled one other $100 on prime of that.
Unfortunately, it appears like that pricing development will proceed. As the final a part of the unique tweet states, “do NOT expect a lower MSRP.” Hopefully that signifies that we at the least will not see even increased pricing for discrete GPUs, nevertheless it does current a chance for a competitor (like AMD or even perhaps Intel) to slide in with decrease pricing and steal some market share. We’ll see what occurs later this yr when the higher-end Arc playing cards and AMD’s Radeon RX 7000 sequence hit the market.